libffi Bindings for Scala Native

Scala Native's C foreign function interface, while very simple to use, still lacks a few key features. This library provides an alternative to the built-in FFI by using libffi — the same underlying library used by JNA.

The only reason you might want to use this is if you need to bind to library functions that receive struct arguments by value, or return a struct by value. Presumably, the need for this library will eventually go away.

Usage

In a Scala Native project, add a resolver to the repository where this library is published, and add a %%% dependency:

resolvers += Resolver.bintrayRepo("nadavwr", "maven"),
libraryDependencies += "com.github.nadavwr" %%% "libffi-scala-native" % "0.5.0"

Every parameter type must have an instance of FfiType[_] in implicit scope. Primitive instances are already defined, so in practice you will only need to do this for structs.

For example, to match libc's div_t struct

typedef struct {
	int quot;		/* quotient */
	int rem;		/* remainder */
} div_t;

we will need to define FfiType[DivT] as follows:

import com.github.nadavwr.ffi._
  
type DivT = CStruct2[CInt, CInt]
implicit class DivOps(private val ptr: Ptr[DivT]) extends AnyVal {
  def quot: CInt = !ptr._1
  def rem: CInt = !ptr._2
}
  
implicit val ffiTypeOfDivT = FfiType.struct[DivT]("div_t", FfiType[CInt], FfiType[CInt])

We can then define bindings for libc's div function, which returns a struct:

val libc = Module.open("libc.dylib") // "libc.so" for Linux
val div = libc.prepare[CInt, CInt, DivT]("div")
val num = stackalloc[CInt]; !num = 10
val denom = stackalloc[CInt]; !denom = 4
val result = stackalloc[DivT]
div(num, denom)(result) // last argument points to result
  
println(s"div(10, 4) = ${result.quot} (${result.rem})")
assert(result.quot == 2)
assert(result.rem == 2)
  
// alternative "inline" allocation style
val result2 = div(num, denom)(stackalloc[DivT])

Limitations

• The return value and all arguments must be preallocated, and provided to bound functions as pointers.
• There is no facility to unload shared libraries once they've been loaded.
• Expect a performance drop typical of libffi. Depending on your needs, this may be negligible.